In data networks or packet-based networks, such as IP (Internet Protocol) based Internet, so-called single shortest-path routing methods are at present the ones most used. With these methods, e.g. OSPF (Open Shortest Path First) and IS-IS (Intermediate System—Intermediate System), link costs are allocated to the links of a network formed by links. This is also referred to as a metric. With this metric, the least expensive or shortest path between two points or nodes is determined. This shortest path is the path with the least accumulated link costs for the links constituting the path. As a very conservative choice, the link cost for each link is regarded as the same (e.g. equal to one). Thus, the shortest path is the path with the lowest number of hops or links.
It is desirable to load all the links in a network equally if possible. An even loading in the network maximizes the tolerance of dynamic load changes, that can occur due both to additional traffic volume and to the failure of links. Methods are known whereby, with a high programming and computing cost, costs are allocated to the links in the network such that optimum routing is achieved with respect to the traffic distribution. These methods are, however, usually so expensive that a distributed implementation in all routers of the network cannot be justified. At present, therefore, non-optimum link costs are mainly used, e.g. the same costs for all links or nodes that are in reverse proportion to the bandwidth of the link. Unavoidable adjustments of link costs are frequently carried out manually, thus carrying a high risk of error. Therefore, a sub-optimum traffic distribution is usually accepted in practical networks.